SizeNet: Weakly Supervised Learning of Visual Size and Fit in Fashion Images

Abstract

Finding clothes that fit is a hot topic in the e-commerce fashion industry. Most approaches addressing this problem are based on statistical methods relying on historical data of articles purchased and returned to the store. Such approaches suffer from the cold start problem for the thousands of articles appearing on the shopping platforms every day, for which no prior purchase history is available. We propose to employ visual data to infer size and fit characteristics of fashion articles. We introduce SizeNet, a weaklysupervised teacher-student training framework that leverages the power of statistical models combined with the rich visual information from article images to learn visual cues for size and fit characteristics, capable of tackling the challenging cold start problem. Detailed experiments are performed on thousands of textile garments, including dresses, trousers, knitwear, tops, etc. from hundreds of different brands.

SizeFlags: Reducing Size and Fit Related Returns in Fashion E-Commerce

“…However the customer body shape estimation suffers from the same personal data requirement as the works mentioned above. Relaxing the constraint on having personal customer data, [17] proposes to focus on articles only and suggests using article images to predict how likely a given article is to have a general size issue.…”

“…As we will discuss below, prior information can come from various sources of different complexity to help shape the prior Beta distribution. Among the more sophisticated ways, we will discuss obtaining prior information about the fit of an article before observing any return using advanced computer vision deep learning models, such as [17], to process the images of the article.…”

“…This value is unknown, but we know that the higher the number of article orders 𝑛, the more confident we are that 𝑠𝑟𝑟 (𝑎) is close to 𝑠𝑟𝑟 𝑡𝑟𝑢𝑒 (𝑎): 𝑠𝑟𝑟 𝑡𝑟𝑢𝑒 (𝑎) = lim 𝑛→∞ 𝑠𝑟𝑟 (𝑎). Similarly to [17], we consider two main elements:…”

“…The article 𝑎 could be considered to have a size issue if 𝑠𝑟𝑟 𝑡𝑟𝑢𝑒 (𝑎) > 𝜋 + 𝜎 when the return rates are calculated over the same articles ordered period of time. Thus, as in [17], we use a binomial law to assess the confidence in the class prediction. The probability of observing 𝑘 size-related returns over a total of 𝑛 orders for that article 𝑎 is modelled by the binomial likelihood…”

Meta-Learning for Size and Fit Recommendation in Fashion

“…Tackling this challenge presents a huge opportunity not only to positively impact customer satisfaction, but also to reduce the environmental footprint, and drive business profitability. A good example is SizeNet [2], which uses customer feedback and product images to predict and localize product-related size issues to warn customers.…”